RE: Trial optimization across multiple endpoints
Title: Block versus diagonal omega
Brian, Let me risk ridicule and mention my favorite algorithm, Genetic algorithm. We've actually done some work in this area (just a little, not published anything). But, it is pretty easy to set up an optimization with constraints. Basically, you simulate (100's of times) whatever scenario you want (we've just done sampling times for pk models), run nonmem (or whatever analysis you want) and look at the expected outcome (as I recall, probability of convergence, probability of covariance step and bias of parameters). Then you also can add a penalty for failing to meet some criteria (in out case, missing Cmax). Very computer intensive (100's-1000's of scenarios, 100's of simulations for each, each with NONMEM run), but can be made as realistic as you want, whatever stats you want, constraints of max exposure, constraints of sample scenarios [not after 1 AM not before 7 AM etc etc) There also is an area of multi objective Genetic algorithm, which might work as well, but that's another story. Mark Mark Sale MD Next Level Solutions, LLC www.NextLevelSolns.com 919-846-9185 A carbon-neutral company See our real time solar energy production at: http://enlighten.enphaseenergy.com/public/systems/aSDz2458
Quoted reply history
-------- Original Message --------
Subject: [NMusers] Trial optimization across multiple endpoints
From: "Corrigan, Brian (Clin Pharm)" < [email protected] >
Date: Sat, August 28, 2010 8:55 pm
To: "nmusers" < [email protected] >
Folks: I knew that 15 plus years of NMUSER messages in my inbox would pay-off someday................. I am asking for you help with a question from a non-US health authority for an upcoming meeting.....not really a NONMEM question, but more of a general modelng question. We have proposed some trial design and optimization work (comparing various trial durations, designs, etc) based on a primary endpoint (ADAS-cog) used universally in the longitudinal disease progression trials for this indication (mild to moderate AD). The health authority opinion is that the trial optimization results could be misleading, as they are based only on one endpoint, and other endpoints (including a co-primary) would need to be taken into account in the overall assesment. My questions are 1) Does anybody else want to go to the meeting in my place? 2) Assuming I get no positive responses to question 1 above, is anyone aware of any model base trial simulations and optimization activities that have utilized more than one endpoint simultaneously? Any that look at two endpoints and weight them in some sort of utility index, or any that just even look at two endpoints and compare the trial simulation results at the end? Much thanks, Brian Corrigan, PhD Pfizer Global Research and Development 50 Pequot Avenue New London, CT 06320 860-732-9189